Hourly Blood Urea Nitrogen Patterns Reveal Hidden ICU Death Risk in Bleeding Patients

Every intensive care unit tracks blood urea nitrogen, a routine laboratory value that most clinicians glance at once and then move past. A new multicenter study suggests that this familiar number may be far more informative than anyone realized, provided it is watched the way intensivists watch heart rates and blood pressures: continuously, hour by hour. The research, published in Advances in Therapy, analyzed nearly six thousand critically ill patients admitted with gastrointestinal bleeding and found that the shape of a patient’s blood urea nitrogen trajectory during the first twenty-four hours in the ICU carries powerful prognostic information that a single admission measurement simply cannot capture.

The study, led by Bin Liu and Chungen Xing of the Second Affiliated Hospital of Soochow University together with colleagues at Chengdu Medical College-affiliated hospitals, assembled a cohort of 5,847 adult patients with gastrointestinal bleeding drawn from four publicly available critical care databases spanning 2008 to 2022: the MIMIC-IV database, the eICU Collaborative Research Database, the Northwestern Medicine ICU database, and the Salzburg Intensive Care database. All four repositories are hosted on the PhysioNet platform and contain de-identified electronic health record data released under institutional review board approval. To be included, patients needed at least twelve blood urea nitrogen observations during the first day of their ICU stay, a demanding requirement that ensured the researchers could map laboratory values onto an hourly grid and reconstruct genuine temporal patterns rather than isolated snapshots.

The cohort was seriously ill. ICU mortality ranged from 18.0 percent to 19.7 percent across the four databases, meaning roughly one in five patients died during their intensive care stay. Men made up between 56.5 and 58.7 percent of each cohort, with women comprising the remaining 41.3 to 43.5 percent. Gastrointestinal bleeding is one of the most common emergencies that brings patients into intensive care, and mortality prediction in this population has long relied on static scoring systems such as the Glasgow-Blatchford Score, the AIMS65 score, and the Rockall score, all of which incorporate admission laboratory values but cannot respond to how a patient evolves after arrival.

The analytical approach was deliberately layered. Generalized additive models, a flexible statistical framework capable of capturing nonlinear relationships without forcing them into rigid parametric shapes, were used to characterize how the concentration and duration of blood urea nitrogen abnormalities jointly related to outcomes. Adjusted logistic regression then evaluated fifteen distinct blood urea nitrogen ranges, allowing the team to ask not just whether high values were dangerous but exactly which thresholds mattered and for how long they needed to persist. Restricted cubic splines examined time-weighted blood urea nitrogen exposure, integrating both magnitude and persistence into a single measure. Finally, the researchers ran meta-analyses across the four databases, performed sensitivity analyses restricted to patients with denser measurement sets, and formally compared static models built on admission values alone against dynamic models that incorporated the full hourly trajectory.

The headline finding is strikingly simple. Patients whose blood urea nitrogen stayed below 30 milligrams per deciliter for more than twelve hours had substantially lower odds of ICU death, with odds ratios ranging from 0.40 to 0.62 across the databases. In other words, keeping this value in the low range through the critical first day was associated with roughly a forty to sixty percent reduction in the odds of mortality compared with patients whose values did not remain in that zone. At the opposite extreme, blood urea nitrogen of 50 milligrams per deciliter or higher sustained for more than eighteen hours markedly increased mortality odds, with odds ratios between 1.55 and 3.39. The dose-response logic is intuitive: the longer a patient remained in the dangerous range, the greater the accumulated risk.

The time-weighted analysis refined this picture further. Prolonged time spent within the 5 to 30 milligrams per deciliter band was associated with lower mortality odds, while greater cumulative exposure to the 40 to 70 milligrams per deciliter range increased risk. Crucially, the associations were nonlinear, meaning the relationship between urea nitrogen exposure and death was not a straight line but a curve with thresholds and inflection points. Equally important for the credibility of the findings, the meta-analysis detected no between-database heterogeneity, and the results remained broadly consistent when the analysis was repeated in subsets of patients with more densely sampled laboratory measurements, addressing the concern that sparse sampling might have distorted the hourly trajectories.

Blood urea nitrogen is a deceptively rich biomarker. The liver produces ammonia during protein metabolism, the urea cycle converts it into urea, and the kidneys excrete it, so blood levels rise with kidney dysfunction, with dehydration and reduced renal perfusion, and with the digestion of blood itself inside the gastrointestinal tract, since digested hemoglobin delivers a large protein load to the gut. In a bleeding patient, an elevated value therefore weaves together several ominous threads at once: hypovolemia, impaired organ perfusion, possible acute kidney injury, and ongoing or severe blood loss. Prior research has already linked elevated blood urea nitrogen to mortality in general ICU populations, in sepsis, in acute heart failure with cardiorenal syndrome, and in acute pancreatitis, and ratios combining urea nitrogen with albumin or creatinine have been proposed as prognostic tools across a range of conditions.

What this study adds is the temporal dimension. A single admission value tells you where a patient started; a trajectory tells you where they are heading. When the researchers compared static models built on admission blood urea nitrogen alone against dynamic models incorporating the hourly trajectory data, the improvement was measurable and statistically significant. The area under the receiver operating characteristic curve, a standard measure of discrimination, rose from 0.678 to 0.701 in the static models to between 0.718 and 0.748 in the dynamic models, with all differences reaching significance on DeLong testing at P values of 0.05 or below. The dynamic models also showed improved reclassification of patients into risk categories and lower Brier scores, indicating better-calibrated overall predictions rather than merely better ranking.

The clinical implications are immediate and practical. Blood urea nitrogen is already drawn repeatedly in virtually every ICU patient, so no new test, no new equipment, and no additional cost would be required to implement trajectory-based monitoring. What would change is interpretation: instead of treating each laboratory result as an isolated data point, early warning protocols could flag patients whose values climb into or persist within the higher ranges during the first day, prompting earlier endoscopic evaluation, more aggressive resuscitation, closer renal monitoring, or escalation of care. Conversely, patients whose values settle and remain below 30 milligrams per deciliter for more than half a day might be identified as lower risk, potentially guiding earlier de-escalation and better allocation of scarce intensive care resources.

The authors are careful about the limits of their design. This was a retrospective cohort study, so the associations describe risk stratification rather than proving that any particular intervention that lowers blood urea nitrogen would improve survival. The requirement for at least twelve measurements in the first twenty-four hours selects a monitored subset of patients, and although sensitivity analyses in denser-measurement subsets were reassuring, the findings will need prospective validation before they reshape guidelines. The datasets came from de-identified records under PhysioNet data use agreements, the study was reported following the TRIPOD statement for prediction research, and the authors declared no conflicts of interest. Even with those caveats, the message is compelling: one of the cheapest, oldest tests in medicine, read as a moving curve rather than a single number, may help clinicians see which bleeding patients are quietly sliding toward death while there is still time to intervene.

Subject of Research: Hourly blood urea nitrogen trajectories as predictors of ICU mortality in critically ill patients with gastrointestinal bleeding

Article Title: Hourly Blood Urea Nitrogen Trajectories and ICU Mortality Risk in Critically Ill Patients with Gastrointestinal Bleeding: A Multicenter Retrospective Cohort Study

Article References: Liu, B., Zheng, X., Xu, Z., Yan, H., Du, Y., Yong, Z., & Xing, C. (2026). Hourly Blood Urea Nitrogen Trajectories and ICU Mortality Risk in Critically Ill Patients with Gastrointestinal Bleeding: A Multicenter Retrospective Cohort Study. Advances in Therapy. https://doi.org/10.1007/s12325-026-03796-w

Image Credits: AI Generated

DOI: 10.1007/s12325-026-03796-w

Keywords: blood urea nitrogen, gastrointestinal bleeding, ICU mortality, risk stratification, dynamic trajectories, critical care, MIMIC-IV, eICU, retrospective cohort study, predictive modeling, acute kidney injury, biomarkers